151. Bagged super wavelets reduction for boosted prostate cancer classification of seldi-tof mass spectral serum profiles
- Author
-
Timothy Hancock, Yvette Everingham, Danny Coomans, and David Donald
- Subjects
business.industry ,Process Chemistry and Technology ,Feature extraction ,Pattern recognition ,computer.software_genre ,medicine.disease ,Computer Science Applications ,Analytical Chemistry ,Random forest ,Reduction (complexity) ,Prostate cancer ,Wavelet ,Coiflet ,Test set ,medicine ,Data mining ,Artificial intelligence ,Linear combination ,business ,computer ,Spectroscopy ,Software ,Mathematics - Abstract
Wavelet based analysis for mass spectrometry (MS) profiles of three groups of patients are analyzed for the purpose of developing a classification model. The first step in our model uses a DWT for feature extraction, using a linear combination of Symlets, Daubechies and Coiflets wavelet bases – collectively known as a super wavelet. Random Forests and Treeboost are then used to analyze the super wavelet coefficients to form the classification model. The method is illustrated using the publicly available prostate SELDI-TOF MS data from the American National Cancer Institute (NCI). The NCI data consists of 322 MS profiles with 15154 M / Z ratios, comprising of 69 malignant, 190 benign and 63 control patients, which we randomly divided into 70% training and 30% testing. From the Random Forest models, the super wavelet performed 2.7% to 5.7% better than other single wavelet types to give a 100% test set prediction rate for cancerous patients.
- Published
- 2006
- Full Text
- View/download PDF